Fix invisible cursor for games using D3D9 software cursors

[Night1099]

Summary:
Games that draw their cursor as a D3D9 sprite and call SetCursor(NULL) to hide the OS cursor
end up with no visible cursor under Remix — the path tracer drops the sprite, and the OS cursor
is actively hidden.

Changes:

• Server (d3d9_cursor.cpp): ShowCursor(TRUE) no longer calls ::SetCursor(nullptr) when
  m_hCursor is NULL (game never called SetCursorProperties), allowing the OS cursor to remain visible
• Bridge client (di_hook.cpp): Hooks SetCursor alongside existing SetCursorPos/GetCursorPos
  hooks. When SetCursor(NULL) is called while the game's D3D9 ShowCursor state is TRUE (menu/UI),
  substitutes a fallback cursor. During gameplay (ShowCursor FALSE), NULL passes through for mouse-look
• Bridge client (d3d9_device.cpp): When the game calls SetCursorProperties with a cursor bitmap,
  creates an HCURSOR on the client side so the hook uses the game's actual cursor icon instead of a
  generic arrow

Configuration:
client.forceSoftwareCursorVisibility (default: true) — set to false to disable

Tested:

• Mount & Blade: Warband — cursor visible in menus, hidden during gameplay
• Stronghold 2 — cursor visible with dynamic icon updates (build, attack, select cursors)
• Games using only SetCursorProperties (hardware cursor) unaffected
• Toggling client.forceSoftwareCursorVisibility = false disables the behavior
• Fullscreen vs windowed mode

[AlexDunn]

While this WAR may be useful, I wonder why it's necessary.

Doesnt DXVK already handle this for us? If not, it should, and would potentially be more robust than adding in another hook (generally we want to support the minimal number of hooks as possible).

[Night1099]

Well upstream dxvk does handle it, the bug is introduced by the bridge

when DXVK's D3D9Cursor::ShowCursor calls ::SetCursor(), it happens in the same process as the game, so it "wins" over whatever the game does.
In dxvk-remix's bridge architecture, the server-side ::SetCursor() call happens in a different process from the game. The game's own ::SetCursor(NULL) call on the client side just overwrites it.

The server-side fix in d3d9_cursor.cpp is the DXVK-level fix — it prevents ShowCursor(TRUE) from nulling the cursor when no hardware cursor exists. However, games using software cursors also call ::SetCursor(NULL) directly (bypassing D3D9), and because of the bridge architecture those calls happen in the client process where DXVK's server-side code can't intercept them. The SetCursor hook in di_hook.cpp covers that path. The existing di_hook infrastructure already hooks other cursor functions (GetCursorPos, SetCursorPos), so the SetCursor hook follows the same pattern.